A transformer voiceprint sample generation method, device, equipment and medium

Abstract

The present application relates to a kind of transformer voiceprint sample generation method and device, including using the transformer mechanism simulation model of target transformer of pre-construction to generate physical field simulation sample, using the audio waveform gradient WGTA diffusion model of target transformer of pre-construction to generate physical field simulation sample, based on the physical field simulation sample and the WGTA diffusion sample, after training variational auto-encoder, generate the voiceprint sample of the target transformer.The present application realizes in the case where guaranteeing the authenticity of voiceprint sample, improve the diversity and complexity of voiceprint sample.The present application also relates to a kind of equipment and storage medium.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method, apparatus, equipment and medium for generating transformer acoustic signature samples. Background Technology

[0002] Acoustic fingerprint samples play a crucial role in transformer fault diagnosis. As a non-invasive monitoring method, acoustic fingerprint samples can capture subtle changes in the internal operating state of a transformer, providing strong support for early fault detection and maintenance. However, obtaining transformer acoustic fingerprint samples is currently quite difficult. First, because transformers generally operate in complex electrical environments, collecting acoustic fingerprint samples carries certain risks, typically requiring specialized personnel and equipment, which significantly increases the cost and risk of acoustic fingerprint sample collection. Second, the sound signals inside a transformer are affected by various factors, such as load changes and environmental noise, which may lead to signal mixing and distortion, affecting the accuracy and reliability of fault diagnosis.

[0003] Current methods for generating transformer acoustic signature samples have serious limitations. First, existing methods are mainly based on simulation data or small-scale experimental data, making it difficult to realistically simulate the complex internal working conditions of transformers. Second, the fault samples generated by current methods rely on fault samples in the training dataset, resulting in overly simplistic content and a lack of diverse and complex real-world fault scenarios. Summary of the Invention

[0004] To address the problem that current methods generate acoustic signature samples with overly simplistic content that fails to reflect real-world fault conditions, this invention provides a method, apparatus, device, and medium for generating transformer acoustic signature samples.

[0005] In a first aspect, the present invention provides a method for generating transformer acoustic signature samples, the method comprising:

[0006] Physical field simulation samples are generated using a pre-constructed transformer mechanism simulation model of the target transformer. The transformer mechanism simulation model is constructed based on the acoustic parameters, physical mechanism characteristics, and prior constraints of the transformer mechanism simulation model of the target transformer.

[0007] Physical field simulation samples are generated using a pre-constructed WGTA diffusion model of the audio waveform gradient of the target transformer. The WGTA diffusion model is obtained by training the physical field simulation samples.

[0008] Based on the physical field simulation samples and the WGTA diffusion samples, the variational autogenerator is trained to generate the acoustic signature samples of the target transformer.

[0009] Based on the above technical solution, the method further includes:

[0010] Using the acoustic and physical parameters of the target transformer, a priori constraints are constructed for the transformer mechanism simulation model of the target transformer. The prior constraints of the transformer mechanism calculation simulation model include the acoustic vibration equation and acoustic boundary conditions.

[0011] The acoustic vibration equation is as follows:

[0012]

[0013]

[0014] Where u is the sound wave vibration displacement vector, t is time, B / A is the nonlinear parameter of the liquid, c is the sound wave vibration propagation velocity, x, y, and z are position variables, and s is the position variable. γ is the specific heat ratio of the gas;

[0015] The acoustic boundary conditions It is the sound pressure gradient along the normal direction, α is the sound absorption coefficient, and P is the sound pressure.

[0016] The geometric model of the target transformer is divided into finite element meshes, and the acoustic pressure field function P(x,y,z,t)=ΣΦ is constructed. i (x,y,z,t)*A i Harmony and acoustic wave equation Where Φ i (x,y,z,t) are Bessel basis functions, A i These are the weighting coefficients;

[0017] Based on the prior constraints of the transformer mechanism simulation model of the target transformer, the acoustic pressure field function and acoustic wave equation of the target transformer are solved by finite element analysis to obtain the transformer mechanism simulation model.

[0018] Based on the above technical solution, further, the step of generating physical field simulation samples using a pre-constructed transformer mechanism simulation model of the target transformer specifically includes:

[0019] The actual physical parameters of the target transformer under different faults are used as simulation mechanism parameters and input into the transformer mechanism simulation model to obtain the sound field data of the target transformer under different faults. After preprocessing the sound field data, the physical field simulation sample is obtained.

[0020] Based on the above technical solution, the method further includes:

[0021] Establish the WGTA acoustic signature diffusion model;

[0022] After classifying the physical field simulation samples according to the corresponding fault types of the target transformer, the physical field simulation samples in each fault type are sorted according to the fault stage to obtain the training set sample distribution function q(x), where x is the sample category and t = 1, 2, ..., T is the fault stage number;

[0023] Then from the fault stage x in step t-1 t-1 up to the fault stage x at step t t The positive transition probability q(x) t |x t-1 );

[0024]

[0025] Where N(·) is the normal distribution function with a mean of 1 / 2. The variance is β t I, where β1,β2,…,β T The coefficient for the noise intensity added at each fault stage;

[0026] Finally, we obtain the values ​​from the initial stage x0 to the t-th stage x for each fault type. t positive transition probability function

[0027] Calculate the fault stage x from step t for each fault type. t Fault stage x at step t-1 t-1 reverse transition probability

[0028] Based on the forward transition probability function and the reverse transition probability, a variational loss function is constructed.

[0029] Where, α t =1-β t ,

[0030] ε t These are the coefficients for each fault stage, θ is the transition probability parameter, and ε is the coefficient. θ It is a probability parameter;

[0031] The WGTA diffusion model is trained until a preset condition is met, resulting in the trained WGTA diffusion model.

[0032] Based on the above technical solution, further, training the WGTA diffusion model until a preset condition is met to obtain the trained WGTA diffusion model specifically includes:

[0033] The loss value calculated by the variational loss function is input into the iterative function. In the process, the result value of the iterative function is obtained, and it is determined whether the result value satisfies the preset convergence condition. If so, the trained WGTA diffusion model is obtained.

[0034] Based on the above technical solution, further, the step of generating WGTA diffusion samples using a pre-constructed audio waveform gradient WGTA diffusion model of the target transformer specifically includes:

[0035] Construct an initial sample that follows a Gaussian distribution, wherein the initial sample x N ~N(0,I);

[0036] The WGTA diffusion sample is generated by inputting the initial sample into the WGTA diffusion model.

[0037] Based on the above technical solution, further, the step of training the variational autogenerator based on the physical field simulation samples and the WGTA diffusion samples to generate the acoustic signature samples of the target transformer specifically includes:

[0038] Construct a hybrid sample dataset consisting of the physical field simulation samples and the WGTA diffusion samples;

[0039] A variational autocoder is constructed to map the voiceprint samples in the mixed sample dataset to latent points in the latent space. The distribution of the latent points is calculated using the minimum error function to determine whether it satisfies a preset normal distribution. If so, the decoder generates the voiceprint samples of the target transformer based on the latent points in the latent space.

[0040] Secondly, the present invention also provides a transformer acoustic signature sample generation device, the device comprising:

[0041] The first construction module is used to generate physical field simulation samples using a pre-constructed transformer mechanism simulation model of the target transformer. The transformer mechanism simulation model is constructed based on the acoustic parameters, physical mechanism characteristics, and prior constraints of the transformer mechanism simulation model of the target transformer.

[0042] The second construction module is used to generate physical field simulation samples using a pre-built audio waveform gradient WGTA diffusion model of the target transformer. The WGTA diffusion model is obtained by training the physical field simulation samples.

[0043] The generation module is used to generate acoustic signature samples of the target transformer after training the variational autoprogrammer based on the physical field simulation samples and the WGTA diffusion samples.

[0044] Based on the above embodiments, the first building module is further configured to:

[0045] Using the acoustic and physical parameters of the target transformer, a priori constraints are constructed for the transformer mechanism simulation model of the target transformer. The prior constraints of the transformer mechanism calculation simulation model include the acoustic vibration equation and acoustic boundary conditions.

[0046] The acoustic vibration equation is as follows:

[0047]

[0048]

[0049] Where u is the sound wave vibration displacement vector, t is time, B / A is the nonlinear parameter of the liquid, c is the sound wave vibration propagation velocity, x, y, and z are position variables, and s is the position variable. γ is the specific heat ratio of the gas;

[0050] The acoustic boundary conditions It is the sound pressure gradient along the normal direction, α is the sound absorption coefficient, and P is the sound pressure.

[0051] The geometric model of the target transformer is divided into finite element meshes, and the acoustic pressure field function P(x,y,z,t)=ΣΦ is constructed. i (x,y,z,t)*A i Harmony and acoustic wave equation Where Φ i (x,y,z,t) are Bessel basis functions, A i These are the weighting coefficients;

[0052] Based on the prior constraints of the transformer mechanism simulation model of the target transformer, the acoustic pressure field function and acoustic wave equation of the target transformer are solved by finite element analysis to obtain the transformer mechanism simulation model.

[0053] Based on the above embodiments, the first building module is further specifically used for:

[0054] The actual physical parameters of the target transformer under different faults are used as simulation mechanism parameters and input into the transformer mechanism simulation model to obtain the sound field data of the target transformer under different faults. After preprocessing the sound field data, the physical field simulation sample is obtained.

[0055] Based on the above embodiments, the second building module is further configured to:

[0056] Establish the WGTA acoustic signature diffusion model;

[0057] After classifying the physical field simulation samples according to the corresponding fault types of the target transformer, the physical field simulation samples in each fault type are sorted according to the fault stage to obtain the training set sample distribution function q(x), where x is the sample category and t = 1, 2, ..., T is the fault stage number;

[0058] Then from the fault stage x in step t-1 t-1 up to the fault stage x at step t t The positive transition probability q(x) t |x t-1 );

[0059]

[0060] Where N(·) is the normal distribution function with a mean of 1 / 2. The variance is β t I, where β1,β2,…,β T The coefficient for the noise intensity added at each fault stage;

[0061] Finally, we obtain the values ​​from the initial stage x0 to the t-th stage x for each fault type. t positive transition probability function

[0062] Calculate the fault stage x from step t for each fault type. t Fault stage x at step t-1 t-1 reverse transition probability

[0063] Based on the forward transition probability function and the reverse transition probability, a variational loss function is constructed.

[0064] Where, α t =1-β t ,

[0065] ε t These are the coefficients for each fault stage, θ is the transition probability parameter, and ε is the coefficient. θ It is a probability parameter;

[0066] The WGTA diffusion model is trained until a preset condition is met, resulting in the trained WGTA diffusion model.

[0067] Based on the above embodiments, the second building module is further specifically used for:

[0068] The loss value calculated by the variational loss function is input into the iterative function. In the process, the result value of the iterative function is obtained, and it is determined whether the result value satisfies the preset convergence condition. If so, the trained WGTA diffusion model is obtained.

[0069] Based on the above embodiments, the second building module is further specifically used for:

[0070] Construct an initial sample that follows a Gaussian distribution, wherein the initial sample x N ~N(0,I);

[0071] The WGTA diffusion sample is generated by inputting the initial sample into the WGTA diffusion model.

[0072] Based on the above embodiments, the generation module is further specifically used for:

[0073] Construct a hybrid sample dataset consisting of the physical field simulation samples and the WGTA diffusion samples;

[0074] A variational autocoder is constructed to map the voiceprint samples in the mixed sample dataset to latent points in the latent space. The distribution of the latent points is calculated using the minimum error function to determine whether it satisfies a preset normal distribution. If so, the decoder generates the voiceprint samples of the target transformer based on the latent points in the latent space.

[0075] Thirdly, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the transformer acoustic signature generation method described in any one of the first aspects.

[0076] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a transformer acoustic signature generation method as described in any one of the first aspects.

[0077] This invention provides a method and apparatus for generating transformer acoustic signature samples, comprising generating physical field simulation samples using a pre-constructed transformer mechanism simulation model of the target transformer, generating physical field simulation samples using a pre-constructed audio waveform gradient WGTA diffusion model of the target transformer, and generating acoustic signature samples of the target transformer after training a variational autogenerator based on the physical field simulation samples and the WGTA diffusion samples. This invention achieves simulation analysis and calculation of the transformer's physical field. By introducing mechanistic knowledge into the construction of the physical field, the internal structure of the transformer can be realistically simulated, the working principle of the transformer can be accurately reproduced, and realistic and reliable acoustic signature simulation samples can be generated. Secondly, to increase the diversity of acoustic signature samples, a transformer acoustic signature generation method based on the Wave Grad diffusion model (Wave Grad for Transformer Audio, WGTA) is proposed. By introducing the Wave Grad diffusion model to describe the fault probability distribution of the acoustic signature samples, diffusion samples under different faults are generated, enhancing the diversity and complexity of the acoustic signature diffusion samples. Finally, a sample fusion method based on variational autoencoder is proposed to combine the generated voiceprint simulation samples and voiceprint diffusion samples, thereby improving the diversity and complexity of voiceprint samples while ensuring their authenticity. Attached Figure Description

[0078] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0079] Figure 1 This is a flowchart illustrating a method for generating transformer acoustic signature samples according to an embodiment of the present invention;

[0080] Figure 2 This is a flowchart illustrating the process of establishing a transformer mechanism calculation simulation model in a transformer acoustic signature sample generation method according to another embodiment of the present invention.

[0081] Figure 3 This is a schematic diagram of the physical field simulation sample generation process in a transformer acoustic fingerprint sample generation method provided by an embodiment of the present invention;

[0082] Figure 4 This is a schematic diagram of the process of establishing a WGTA acoustic signature diffusion model in a transformer acoustic signature sample generation method provided in another embodiment of the present invention;

[0083] Figure 5 This is a schematic diagram of the WGTA diffusion sample generation process in a transformer acoustic signature sample generation method provided in an embodiment of the present invention;

[0084] Figure 6This is a schematic diagram of the process of sample fusion based on variational autoencoder in a transformer acoustic fingerprint sample generation method provided by another embodiment of the present invention;

[0085] Figure 7 This is a schematic diagram of diffusion model transfer in a transformer acoustic signature sample generation method provided in another embodiment of the present invention;

[0086] Figure 8 This is a schematic diagram of a transformer acoustic signature sample generation device provided in another embodiment of the present invention. Detailed Implementation

[0087] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0088] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0089] The following will be combined with the appendix Figure 1 The method for generating transformer acoustic signature samples provided in this embodiment of the invention includes the following steps:

[0090] S1. Generate physical field simulation samples using a pre-constructed transformer mechanism simulation model of the target transformer. The transformer mechanism simulation model is constructed based on the acoustic parameters, physical mechanism characteristics, and prior constraints of the transformer mechanism simulation model of the target transformer.

[0091] S2. Generate physical field simulation samples using a pre-built WGTA diffusion model of the target transformer's audio waveform gradient. The WGTA diffusion model is obtained by training the physical field simulation samples.

[0092] S3. Based on the physical field simulation samples and the WGTA diffusion samples, the variational autogenerator is trained to generate the acoustic signature samples of the target transformer.

[0093] Based on the above embodiments, the method further includes step S1a;

[0094] Step S1a specifically includes:

[0095] S11a. Using the acoustic and physical parameters of the target transformer, construct the prior constraints of the transformer mechanism simulation model of the target transformer. The prior constraints of the transformer mechanism calculation simulation model include the acoustic vibration equation and acoustic boundary conditions.

[0096] The acoustic vibration equation is as follows:

[0097]

[0098]

[0099] Where u is the sound wave vibration displacement vector, t is time, B / A is the nonlinear parameter of the liquid, c is the sound wave vibration propagation velocity, x, y, and z are position variables, and s is the position variable. γ is the specific heat ratio of the gas;

[0100] The acoustic boundary conditions It is the sound pressure gradient along the normal direction, α is the sound absorption coefficient, and P is the sound pressure.

[0101] S12a. Divide the geometric model of the target transformer into a finite element mesh and construct the sound pressure field function P(x,y,z,t)=ΣΦ i (x,y,z,t)*A i Harmony and acoustic wave equation Where Φ i (x,y,z,t) are Bessel basis functions, A i These are the weighting coefficients;

[0102] S13a. Based on the prior constraints of the transformer mechanism simulation model of the target transformer, the acoustic pressure field function and acoustic wave equation of the target transformer are solved by finite element analysis to obtain the transformer mechanism simulation model.

[0103] Based on the above embodiments, step S1 further includes:

[0104] S14. Input the actual physical parameters of the target transformer under different faults into the transformer mechanism simulation model as simulation mechanism parameters to obtain the sound field data of the target transformer under different faults. After preprocessing the sound field data, obtain the physical field simulation sample.

[0105] Based on the above embodiments, the method further includes step S2a, which specifically includes:

[0106] S21a. Establish the WGTA acoustic signature diffusion model;

[0107] S22a. After classifying the physical field simulation samples according to the corresponding fault types of the target transformer, the physical field simulation samples in each fault type are sorted according to the fault stage to obtain the training set sample distribution function q(x), where x is the sample category and t = 1, 2, ..., T is the fault stage number;

[0108] Then from the fault stage x in step t-1 t-1 up to the fault stage x at step t t The positive transition probability q(x) t |x t-1 )

[0109]

[0110] Where N(·) is the normal distribution function with a mean of 1 / 2. The variance is β t I, where β1,β2,…,β T The coefficient for the noise intensity added at each fault stage;

[0111] S23a, Finally, the fault stage x0 from the initial stage x0 to the t-th step x is obtained for each fault type. t positive transition probability function

[0112] S24a. Calculate the fault stage x from step t for each fault type. t Fault stage x at step t-1 t-1 reverse transition probability

[0113] S25a. Based on the forward transition probability function and the reverse transition probability, construct a variational loss function. Where, α t =1-β t , ε t These are the coefficients for each fault stage, θ is the transition probability parameter, and ε is the coefficient. θ It is a probability parameter;

[0114] S26a. Train the WGTA diffusion model until the preset conditions are met to obtain the trained WGTA diffusion model.

[0115] Based on the above embodiments, step S26a further includes:

[0116] The loss value calculated by the variational loss function is input into the iterative function. In the process, the result value of the iterative function is obtained, and it is determined whether the result value satisfies the preset convergence condition. If so, the trained WGTA diffusion model is obtained.

[0117] Based on the above embodiments, step S2 further includes:

[0118] S27. Construct an initial sample that follows a Gaussian distribution, wherein the initial sample x N ~N(0,I);

[0119] S28. After inputting the initial sample into the WGTA diffusion model, the WGTA diffusion sample is generated.

[0120] Based on the above embodiments, step S3 further includes:

[0121] S31. Construct a hybrid sample dataset consisting of the physical field simulation sample and the WGTA diffusion sample;

[0122] S32. Construct a variational autocoder to map the voiceprint samples in the mixed sample dataset to potential points in the latent space. Calculate whether the distribution of the potential points satisfies a preset normal distribution using the minimum error function. If so, the decoder generates the voiceprint samples of the target transformer based on the potential points in the latent space.

[0123] It should be understood that the present invention relates to a method for generating transformer acoustic signature samples, which achieves the following technical effects:

[0124] Firstly, by simulating and analyzing the physical field of the transformer and incorporating mechanistic knowledge into the construction of the physical field, the problem of traditional methods being unable to realistically simulate the complex internal working state of the transformer is solved.

[0125] Secondly, by building a transformer acoustic signature generation model based on the Wave Grad diffusion model, the fault probability distribution of the acoustic signature model is described, and diffusion samples under different faults are generated, thereby enhancing the diversity and complexity of acoustic signature diffusion samples.

[0126] Third, by proposing a sample fusion method based on variational autoencoders, and combining the generated voiceprint simulation samples and voiceprint diffusion samples, the diversity and complexity of voiceprint samples are improved while ensuring the authenticity of the voiceprint samples.

[0127] This invention relates to a method for generating transformer acoustic signature samples, comprising a sample generation strategy based on physical field mechanism calculations, a sample generation strategy based on the WGTA diffusion model, and a sample fusion strategy based on a variational autoencoder. Taking transformer acoustic signature fault diagnosis in the power industry as an example, this invention achieves the above objectives through the following technical solutions:

[0128] The following will be combined with the appendix Figures 2 to 7 The method for generating transformer acoustic signature samples provided in this embodiment of the invention includes the following steps:

[0129] like Figure 2 As shown, the process includes step one: establishing a simulation model for transformer mechanism calculation.

[0130] First, a transformer mechanism simulation model is established using finite element analysis. By introducing mechanism calculations, the working principle of the transformer is accurately reproduced, generating realistic and reliable acoustic signature simulation samples. The specific steps are as follows:

[0131] Step 1: Collect the transformer's acoustic parameters and physical mechanism characteristics. In addition to general acoustic parameters, it is also necessary to collect information on the transformer's acoustic characteristics, such as sound velocity c, density m, sound absorption coefficient α, and specific heat ratio γ of the propagation material.

[0132] Step 2: Construct the prior constraints for the transformer mechanism calculation and simulation model.

[0133] Define the constraints for the equations. Based on the shape and size of the windings, core, and other structures, construct the constraints for the acoustic vibration equations in different media:

[0134]

[0135]

[0136] Where u is the acoustic wave vibration displacement vector, t is time, and B / A is the nonlinear parameter of the liquid.

[0137] Define acoustic boundary conditions. By studying the transformer mechanism and considering the reflection and propagation of sound waves inside the transformer, we can define the acoustic boundary conditions inside the transformer. In practical applications, the boundary conditions are usually a mixture of absorption and reflection. Therefore, let the boundary conditions be:

[0138]

[0139] in Let α represent the sound pressure gradient along the normal direction, α represent the sound absorption coefficient, and P represent the sound pressure. The sound absorption coefficient α can be varied to reflect the absorption and reflection occurring at the boundary.

[0140] Place the sound source in an appropriate location in the model to simulate the generation of the sound source.

[0141] Step 3: Mesh Generation and Mechanism Calculation Analysis. The geometric model of the transformer is divided into appropriate finite element meshes. Mechanism equations for the transformer's sound field and vibration field are constructed to achieve multi-physics mechanism calculations. First, mesh generation is performed. In this embodiment, hexahedrons are used as the basic unit. After determining the mesh density, mesh generation is performed. Finite element analysis is then conducted. A set of orthogonal basis functions is used to describe the sound field distribution; the Bessel function is chosen as the sound field basis function. These basis functions can be expanded throughout the entire acoustic region. The complete representation of the sound field is shown below:

[0142] P(x,y,z,t)=ΣΦ i (x,y,z,t)*A i

[0143] Where P(x,y,z,t) is the sound pressure field, Φ i (x,y,z,t) are Bessel basis functions, A i These are the weighting coefficients.

[0144] Simultaneously, the acoustic wave equation is established:

[0145]

[0146] Step 4: Solving the physical field mechanism equations. Based on finite element analysis, the physical field mechanism equations of the transformer are solved, and the transformer mechanism calculation and simulation model is constructed.

[0147] like Figure 3 As shown, step two is included: generating physical field simulation samples.

[0148] Step 1: Set the simulation parameters for the transformer's physical field. Based on the actual situation, set the simulation parameters for the transformer's physical field to complete the parameter initialization of the transformer mechanism calculation simulation model.

[0149] Step 2: Generation of Physics Simulation Samples. The task of generating acoustic samples is completed by calculating and simulating the transformer mechanism, and the acoustic field data of the model at different time steps is recorded.

[0150] Step 3: Voiceprint Sample Preprocessing. The generated voiceprint simulation samples are analyzed and optimized to produce realistic voiceprint simulation samples.

[0151] like Figure 4 and Figure 7 As shown, step three is: establishing a voiceprint diffusion model based on WGTA.

[0152] By establishing a WGTA acoustic signature diffusion model, the fault probability distribution of acoustic signature samples is described to generate diffusion samples under different fault conditions, thereby enhancing the diversity and complexity of acoustic signature diffusion samples. From noise signal x... T Inverse denoising is performed to generate voiceprint samples x0. To obtain the inverse transition probability p... θ (x t-1 |x t First, the WGTA speaker diffusion model needs to be trained in a forward diffusion manner. The training process is as follows:

[0153] Step 1: Calculate the positive transition probability q(x) t |x t-1 Let the training set sample distribution be q(x0), where x0 is the sample class. Let t = 1, 2, ..., T be the index of each step in the diffusion stage. Then the transition probability at each step is q(x0). t |x t-1 Therefore, the specific expression for the transition probability is:

[0154]

[0155] Where N(·) represents a normal distribution with a mean of 1 / 2. The variance is β t I, where β1,β2,…,β T The adjustable variance coefficient is used to control the noise intensity added at each step. Furthermore, the transition distribution at any given time can be derived from the above formula:

[0156]

[0157] Where α t =1-β t ,

[0158] Step 2: Calculate the reverse transition probability p θ (x t-1 |x t Due to the reverse transition probability p θ (x t-1 |x t ) cannot be calculated, therefore q(x) is used. t-1 |x t Approximation is performed using x0).

[0159] set up

[0160] It can be deduced that:

[0161]

[0162] in:

[0163]

[0164]

[0165] Therefore, we can conclude that:

[0166] Step 3: Calculate the variational loss function. The variational loss function can be derived from the KL divergence formula as follows:

[0167]

[0168] Step 4: WGTA diffusion model training. Iterate through the function until convergence:

[0169]

[0170] The training continues until convergence is achieved and the model training is complete.

[0171] like Figure 5 As shown, step four is included: WGTA diffusion sample generation.

[0172] After the WGTA model converges, WGTA diffusion sample generation can be performed. N ~N(0,I)

[0173] Step 1: Initialize the initial noise distribution. First, initialize the samples. The initial samples follow a Gaussian distribution. Let N rounds be passed, where x... N ~N(0,I).

[0174] Step 2: Iteratively generate WGTA diffusion samples. Based on the above model, multiple iterations are performed to generate WGTA diffusion samples.

[0175] like Figure 6 As shown, step five is: sample fusion based on variational autoencoder.

[0176] Step 1: Construct a hybrid sample dataset consisting of simulation samples and diffusion samples.

[0177] Step 2: Construct a variational autoencoder. The encoder maps the voiceprint samples to the mean and variance of the latent space, and the decoder generates voiceprint samples from random points in the latent space.

[0178] Step 3: Perform variational autoencoder training. By minimizing the reconstruction error, ensure that the distribution of latent points approximates a standard normal distribution.

[0179] Step 4: Generate fused samples. New acoustic signature features are sampled from the latent space of the VAE to achieve the fusion of simulated samples and diffused samples.

[0180] This invention relates to a method for generating transformer acoustic signature samples, particularly a sample generation strategy based on physical field mechanism calculations, a sample generation strategy based on the WGTA diffusion model, and a sample fusion strategy based on a variational autoencoder. It enables simulation analysis and calculation of the transformer's physical field. By incorporating mechanistic knowledge into the construction of the physical field, the internal structure of the transformer can be realistically simulated, accurately reproducing the transformer's working principle and generating realistic and reliable acoustic signature simulation samples. Secondly, to increase the diversity of acoustic signature samples, a transformer acoustic signature generation method based on the Wave Grad diffusion model (Wave Grad for Transformer Audio, WGTA) is proposed. This method introduces the Wave Grad diffusion model to describe the fault probability distribution of the acoustic signature samples, generating diffusion samples under different fault conditions, thus enhancing the diversity and complexity of the acoustic signature diffusion samples. Finally, a sample fusion method based on a variational autoencoder is proposed to combine the generated acoustic signature simulation samples and acoustic signature diffusion samples, improving the diversity and complexity of the acoustic signature samples while ensuring their realism.

[0181] The following will be combined with the appendix Figure 8 An embodiment of the present invention provides a transformer acoustic signature sample generation device, the device comprising:

[0182] The first construction module is used to generate physical field simulation samples using a pre-constructed transformer mechanism simulation model of the target transformer. The transformer mechanism simulation model is constructed based on the acoustic parameters, physical mechanism characteristics, and prior constraints of the transformer mechanism simulation model of the target transformer.

[0183] The second construction module is used to generate physical field simulation samples using a pre-built audio waveform gradient WGTA diffusion model of the target transformer. The WGTA diffusion model is obtained by training the physical field simulation samples.

[0184] The generation module is used to generate acoustic signature samples of the target transformer after training the variational autoprogrammer based on the physical field simulation samples and the WGTA diffusion samples.

[0185] Based on the above embodiments, the first building module is further configured to:

[0186] Using the acoustic and physical parameters of the target transformer, a priori constraints are constructed for the transformer mechanism simulation model of the target transformer. The prior constraints of the transformer mechanism calculation simulation model include the acoustic vibration equation and acoustic boundary conditions.

[0187] The acoustic vibration equation is as follows:

[0188]

[0189]

[0190] Where u is the sound wave vibration displacement vector, t is time, B / A is the nonlinear parameter of the liquid, c is the sound wave vibration propagation velocity, x, y, and z are position variables, and s is the position variable. γ is the specific heat ratio of the gas;

[0191] The acoustic boundary conditions It is the sound pressure gradient along the normal direction, α is the sound absorption coefficient, and P is the sound pressure.

[0192] The geometric model of the target transformer is divided into finite element meshes, and the acoustic pressure field function P(x,y,z,t)=ΣΦ is constructed. i (x,y,z,t)*A i Harmony and acoustic wave equation Where Φ i (x,y,z,t) are Bessel basis functions, A i These are the weighting coefficients;

[0193] Based on the prior constraints of the transformer mechanism simulation model of the target transformer, the acoustic pressure field function and acoustic wave equation of the target transformer are solved by finite element analysis to obtain the transformer mechanism simulation model.

[0194] Based on the above embodiments, the first building module is further specifically used for:

[0195] The actual physical parameters of the target transformer under different faults are used as simulation mechanism parameters and input into the transformer mechanism simulation model to obtain the sound field data of the target transformer under different faults. After preprocessing the sound field data, the physical field simulation sample is obtained.

[0196] Based on the above embodiments, the second building module is further configured to:

[0197] Establish the WGTA acoustic signature diffusion model;

[0198] After classifying the physical field simulation samples according to the corresponding fault types of the target transformer, the physical field simulation samples in each fault type are sorted according to the fault stage to obtain the training set sample distribution function q(x), where x is the sample category and t = 1, 2, ..., T is the fault stage number;

[0199] Then from the fault stage x in step t-1 t-1 up to the fault stage x at step t t The positive transition probability q(x) t |x t-1);

[0200]

[0201] Where N(·) is the normal distribution function with a mean of 1 / 2. The variance is β t I, where β1,β2,…,β T The coefficient for the noise intensity added at each fault stage;

[0202] Finally, we obtain the values ​​from the initial stage x0 to the t-th stage x for each fault type. t positive transition probability function

[0203] Calculate the fault stage x from step t for each fault type. t Fault stage x at step t-1 t-1 reverse transition probability

[0204] Based on the forward transition probability function and the reverse transition probability, a variational loss function is constructed.

[0205] Where, α t =1-β t ,

[0206] ε t These are the coefficients for each fault stage, θ is the transition probability parameter, and ε is the coefficient. θ It is a probability parameter;

[0207] The WGTA diffusion model is trained until a preset condition is met, resulting in the trained WGTA diffusion model.

[0208] Based on the above embodiments, the second building module is further specifically used for:

[0209] The loss value calculated by the variational loss function is input into the iterative function. In the process, the result value of the iterative function is obtained, and it is determined whether the result value satisfies the preset convergence condition. If so, the trained WGTA diffusion model is obtained.

[0210] Based on the above embodiments, the second building module is further specifically used for:

[0211] Construct an initial sample that follows a Gaussian distribution, wherein the initial sample x N ~N(0,I);

[0212] The WGTA diffusion sample is generated by inputting the initial sample into the WGTA diffusion model.

[0213] Based on the above embodiments, the generation module is further specifically used for:

[0214] Construct a hybrid sample dataset consisting of the physical field simulation samples and the WGTA diffusion samples;

[0215] A variational autocoder is constructed to map the voiceprint samples in the mixed sample dataset to latent points in the latent space. The distribution of the latent points is calculated using the minimum error function to determine whether it satisfies a preset normal distribution. If so, the decoder generates the voiceprint samples of the target transformer based on the latent points in the latent space.

[0216] This embodiment provides a transformer acoustic signature sample generation device that enables simulation analysis and calculation of the transformer's physical field. By incorporating mechanistic knowledge into the construction of the physical field, the internal structure of the transformer can be realistically simulated, accurately restoring the transformer's working principle and generating realistic and reliable acoustic signature simulation samples. Secondly, to increase the diversity of acoustic signature samples, a transformer acoustic signature generation method based on the Wave Grad diffusion model (Wave Grad for TransformerAudio, WGTA) is proposed. This method introduces the Wave Grad diffusion model to describe the fault probability distribution of the acoustic signature samples, generating diffusion samples under different fault conditions, thus enhancing the diversity and complexity of the acoustic signature diffusion samples. Finally, a sample fusion method based on a variational autoencoder is proposed to combine the generated acoustic signature simulation samples and acoustic signature diffusion samples, improving the diversity and complexity of the acoustic signature samples while ensuring their realism.

[0217] Furthermore, embodiments of the present invention include a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a transformer acoustic signature sample generation method as described in any of the above technical solutions.

[0218] This invention also includes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a transformer acoustic signature generation method as described in any of the above technical solutions.

[0219] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

[0220] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0221] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0222] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0223] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0224] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for generating transformer acoustic signature samples, characterized in that, The method includes: Physical field simulation samples are generated using a pre-constructed transformer mechanism simulation model of the target transformer. The transformer mechanism simulation model is constructed based on the acoustic parameters, physical mechanism characteristics, and prior constraints of the transformer mechanism simulation model of the target transformer. WGTA diffusion samples are generated using a pre-constructed audio waveform gradient WGTA diffusion model of the target transformer, wherein the WGTA diffusion model is obtained by training the physical field simulation samples. Based on the physical field simulation samples and the WGTA diffusion samples, the variational autoencoder is trained to generate the acoustic signature samples of the target transformer. The method further includes: Using the acoustic and physical parameters of the target transformer, a priori constraints for the transformer mechanism simulation model of the target transformer are constructed. The prior constraints for the transformer mechanism simulation model include the acoustic vibration equation and acoustic boundary conditions. The acoustic vibration equation is as follows: in, u It is the displacement vector of the sound wave vibration. It is time. B / A For nonlinear parameters of the liquid, c The speed of sound wave vibration propagation. x, y and z It is a positional variable. s For position variables , γ It is the specific heat ratio of the gas; The acoustic boundary conditions , It is the sound pressure gradient in the normal direction. a It is the sound absorption coefficient. P It's sound pressure; The geometric model of the target transformer is divided into finite element meshes to construct the sound pressure field function. Harmony and acoustic wave equation ,in For Bessel basis functions, These are the weighting coefficients; Based on the prior constraints of the transformer mechanism simulation model of the target transformer, the acoustic pressure field function and acoustic wave equation of the target transformer are solved by finite element analysis to obtain the transformer mechanism simulation model. The generation of physical field simulation samples using a pre-constructed transformer mechanism simulation model of the target transformer specifically includes: The actual physical parameters of the target transformer under different faults are used as simulation mechanism parameters and input into the transformer mechanism simulation model to obtain the sound field data of the target transformer under different faults. After preprocessing the sound field data, the physical field simulation sample is obtained. The method further includes: Establish the WGTA acoustic signature diffusion model; After classifying the physical field simulation samples according to the corresponding fault types of the target transformer, and then sorting the physical field simulation samples in each fault type according to the fault stage, the training set sample distribution function is obtained. q ( x ),in x For sample categories, This refers to the fault stage number; Then from the first t -1 Fault Stage To the t Step Fault Phase positive transition probability ; in, It is a normal distribution function with a mean of 1 / 2. The variance is ,in, The coefficient for the noise intensity added at each fault stage; Finally, the initial stage of each fault type is obtained. up to the t-th fault stage positive transition probability function ; Calculate the fault type from the first t Step Fault Phase To the t -1 Fault Stage reverse transition probability ; Based on the forward transition probability function and the reverse transition probability, a variational loss function is constructed. ,in, , , , , These are the coefficients for each stage of the failure. It is the transition probability parameter. It is a probability parameter; The WGTA diffusion model is trained until a preset condition is met, resulting in the trained WGTA diffusion model.

2. The method according to claim 1, characterized in that, The step of training the WGTA diffusion model until a preset condition is met to obtain the trained WGTA diffusion model specifically includes: The loss value calculated by the variational loss function is input into the iterative function. In the process, the result value of the iterative function is obtained; Determine whether the result value meets the preset convergence condition. If so, the trained WGTA diffusion model is obtained.

3. The method according to claim 1, characterized in that, The generation of WGTA diffusion samples using a pre-built audio waveform gradient WGTA diffusion model of the target transformer specifically includes: Construct an initial sample that follows a Gaussian distribution, the initial sample... ; The WGTA diffusion sample is generated by inputting the initial sample into the WGTA diffusion model.

4. The method according to claim 1, characterized in that, After training the variational autoencoder based on the physical field simulation samples and the WGTA diffusion samples, the acoustic signature samples of the target transformer are generated, specifically including: Construct a hybrid sample dataset consisting of the physical field simulation samples and the WGTA diffusion samples; A variational autoencoder is constructed to map the voiceprint samples in the mixed sample dataset to latent points in the latent space. The distribution of the latent points is calculated using the minimum error function to determine whether it satisfies a preset normal distribution. If so, the decoder generates the voiceprint samples of the target transformer based on the latent points in the latent space.

5. A transformer acoustic signature sample generation device, characterized in that, The device includes: The first construction module is used to generate physical field simulation samples using a pre-constructed transformer mechanism simulation model of the target transformer. The transformer mechanism simulation model is constructed based on the acoustic parameters, physical mechanism characteristics, and prior constraints of the transformer mechanism simulation model of the target transformer. The second construction module is used to generate WGTA diffusion samples using a pre-built audio waveform gradient WGTA diffusion model of the target transformer. The WGTA diffusion model is obtained by training the physical field simulation samples. The generation module is used to generate the acoustic signature sample of the target transformer after training the variational autoencoder based on the physical field simulation sample and the WGTA diffusion sample. The device further includes: The third construction module is used to construct the prior constraints of the transformer mechanism simulation model of the target transformer using the acoustic and physical parameters of the target transformer. The prior constraints of the transformer mechanism simulation model include the acoustic vibration equation and acoustic boundary conditions. The acoustic vibration equation is as follows: in, u It is the displacement vector of the sound wave vibration. It is time. B / A For nonlinear parameters of the liquid, c The speed of sound wave vibration propagation. x, y and z It is a positional variable. s For position variables , γ It is the specific heat ratio of the gas; The acoustic boundary conditions , It is the sound pressure gradient in the normal direction. a It is the sound absorption coefficient. P It's sound pressure; The geometric model of the target transformer is divided into finite element meshes to construct the sound pressure field function. Harmony and acoustic wave equation ,in For Bessel basis functions, These are the weighting coefficients; Based on the prior constraints of the transformer mechanism simulation model of the target transformer, the acoustic pressure field function and acoustic wave equation of the target transformer are solved by finite element analysis to obtain the transformer mechanism simulation model. The generation of physical field simulation samples using a pre-constructed transformer mechanism simulation model of the target transformer specifically includes: The actual physical parameters of the target transformer under different faults are used as simulation mechanism parameters and input into the transformer mechanism simulation model to obtain the sound field data of the target transformer under different faults. After preprocessing the sound field data, the physical field simulation sample is obtained. The device further includes: The fourth building module is used to establish the WGTA acoustic signature diffusion model; After classifying the physical field simulation samples according to the corresponding fault types of the target transformer, and then sorting the physical field simulation samples in each fault type according to the fault stage, the training set sample distribution function is obtained. q ( x ),in x For sample categories, This refers to the fault stage number; Then from the first t -1 Fault Stage To the t Step Fault Phase positive transition probability ; in, It is a normal distribution function with a mean of 1 / 2. The variance is ,in, The coefficient for the noise intensity added at each fault stage; Finally, the initial stage of each fault type is obtained. up to the t-th fault stage positive transition probability function ; Calculate the fault type from the first t Step Fault Phase To the t -1 Fault Stage reverse transition probability ; Based on the forward transition probability function and the reverse transition probability, a variational loss function is constructed. ,in, , , , , These are the coefficients for each stage of the failure. It is the transition probability parameter. It is a probability parameter; The WGTA diffusion model is trained until a preset condition is met, resulting in the trained WGTA diffusion model.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the transformer acoustic signature sample generation method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the transformer acoustic signature sample generation method according to any one of claims 1 to 4.

Images